System Design Interview: Platform Architecture Cheat Sheet
YouTube
YouTube uses a microservices architecture. The video upload service handles transcoding and stores files in object
storage like S3. A metadata service stores video titles, tags, and descriptions in a relational or NoSQL database. User
history and behavior feed into a machine learning-based recommendation system. Video playback is optimized via
CDNs for low latency. Comments, likes, and subscriptions are managed through scalable services using NoSQL for
flexibility. Search is powered by Elasticsearch, and the whole system uses JWT or OAuth2 for secure authentication.
Instagram
Instagram's architecture centers around a user graph, content service, and timeline generator. Users upload
photos/videos, which are processed and stored in object storage. A timeline service fetches content from followed users
and ranks it using ML models. The explore feature uses embedding and clustering to show relevant content. Stories and
reels are separate high-throughput services. Media delivery is handled via CDN, and notifications are sent via
background workers.
WhatsApp
WhatsApp focuses on low latency and security. Messages are encrypted using the Signal Protocol. When a message is
sent, it's placed in a queue and delivered via a fan-out model to all recipients. Message acknowledgements (ticks) show
delivery status. Group chats are managed through replicated queues or topics. Media is uploaded to a secure server
with links shared in the message. Contact syncing and presence are handled by specialized services, and push
notifications keep users updated.
Facebook
Facebook uses a distributed graph database to manage friend relationships. Each user has a feed generated by
aggregating posts from friends, pages, and groups, then ranked by ML. Likes, comments, and reactions are stored
separately to scale independently. The system supports live content, ads, groups, and marketplaces - each as its own
microservice. Data is cached with Memcached or TAO to reduce DB hits. Everything is globally distributed to handle
billions of users.
Swiggy
Swiggy connects customers, restaurants, and delivery partners through real-time systems. When a user places an
System Design Interview: Platform Architecture Cheat Sheet
order, it's sent to the order service, which coordinates with restaurant availability and inventory. A delivery assignment
service uses location-based algorithms to find the nearest delivery partner. Real-time tracking is powered by GPS and
push notifications. Menus and search use caching and indexing. Payments are integrated with third-party gateways and
updates are sent via background queues.
LinkedIn
LinkedIn's core is the user profile and connection graph. The feed is built using content from 1st and 2nd-degree
connections, ranked using relevance and engagement. Jobs, messaging, and notifications are modular services. The
recommendation system powers 'people you may know' and job matches using collaborative filtering and embeddings.
Elasticsearch helps with fast typeahead search. The platform supports a mix of OLTP for real-time interactions and
OLAP for analytics.
